That's what I don't get. It would be so simple to certify for these kinds of things before allowing on public roads. Everyone acts like it would be such an impediment. I have to demonstrate I can see without glasses for my license, otherwise I have to drive with glasses. Why does a self-driving car not have to make a basic demonstration of visual/sensor skills before being put on the road? Since self-driving largely relies on the superiority of its sensors for its safety, there should be much higher expectations for sensor testing than a human could actually perform.
I don't even care about your kind of comment any more, because it's totally fine with me if we turn 'doing dumb things' into a capital offense. But what we can't do is allow companies like Uber, of all companies, to make that decision for us. Any way, we've all done dumb things at one point or another. Most of us would not be here if it came to pass. I certainly wouldn't.
We can know how well humans deal with edge cases by tracking self driving cars and after modelling driving that resembles human driving (meaning mostly diveway to work to grocery store to driveway type stuff). Come up with a criteria for 'difficult event' and log how many occur during a day of normal driving. Compare that statistic to real human accident statistics, extract the gap and there you have the number of accidents prevented by humans.
Human drivers are very safe compared to Uber, that we know. About other self-driving technologies we know nothing, but I really don't think they are that far ahead of Uber.
Musk acts more like these are hobbies for him. He starts making a model car, then he sees a model rocket in the store that he likes and takes it home and starts that despite not finishing the car. See what I did there?
The first thing regulation should enforce is a set of sensors that are capable of handing these types of events then. Uber obviously isn't, who knows if Waymo is?
I call it an edge case because a person dressed in dark standing in a darkly lit section of road with an outcropping behind and choosing to walk into the path of the car is not a common event. Yet it is an event that must be considered, as well as the other million events that contain some novel circumstance.
I've tried a lot of distributions over the years and I can't say I've ever found an interface that felt totally 'light' for linux. I'm not a big fan of OS/X but I can say that it is one thing they got right. Windows even feels light by comparison. I hope it is changing for the better.
So we're going to keep doing this for a million situations? That's a million fatalities. How many edge cases will there be in the real world? Equipment failing is a valid edge case, which brings up my next point. What distance-between interventions misses is the ability of the sensors to operate in all conditions. The camera feeds I saw from the car were terribly dark. Lidar is easily fooled by textured backgrounds. Uber could have avoided this becoming public simply by driving in the day. This tells me not only that it was a weakness that didn't get caught, it was a weakness they didn't even know about.
I'm not talking about accidents though, I'm talking about potentially accident-causing situations. It's a different set of events, because many possible accidents are avoided by humans.
We need self driving vehicles only IF they ever make a dent in the fatality rate. So far I see no evidence at best and I'm not prepared to give them leniency on killing people now based on a mere promise and I don't think anyone should. All evidence right now is that automation isn't coming anywhere close to human safety, which really is quite safe if you take it in perspective of miles driven by humans every day. Being overweight is probably more dangerous and people volunteer for that.
As I said above, 80% of driving is easy, 20% of driving is hard. Not difficult for Waymo to stick to the situations they know they are good at while still being a ticking time bomb on 20% of edge cases. We need full disclosure from these companies, including proof that they can in fact see and recognize ALL objects at ALL times of day and in ALL weather that may be in danger from the car, or be a danger to the car and that they react appropriately.
That sounds great. Why can't corporations get there without killing people on public roads. It's not like there aren't huge testing facilities all over the place. It's not like odd test cases are hard to think up.
80% of driving is easy, 20% of it is full of a million edge cases like this one. Furthermore, edge cases like this are easily avoidable if you have a vested interest in choosing where and when to drive. Waymo could easily rack up 5600 miles and never hit an edge case. We only got elucidated on this one because of a bad test driver and it gave us a window into how bad the situation really is.
This is about saving lives, and is no place for a tit for tat argument. Why can't both sides just try be as safe as FUCKING possible so if one side makes a mistake it is covered by the other side? Is it really that hard to comprehend? Are people really that insensitive that they can't inconvenience themselves by grazing a brake pedal if something unexpected happens up the road?
Another thing I should have added for 2) was the fact that Uber *was* making the requirement for interventions and someone died. So obviously the measurement is bull.
I think that's obvious, but it doesn't change the fact that self driving should be held to a standard where they can accommodate people doing stupid things, otherwise the whole line about saving lines is just blowing smoke. Defensive driving is a 'thing' for good reason and very little of it changes if a computer is driving. If you see something curious, you slow down. Simple?
1) But they have to make something marketable. Then why didn't Uber have a LIDAR array?
2) That's a poor measure. We need to know exactly how many and what type of obstacles the cars are dealing with in those miles. The biggest problem is, self-driving is easy 80% of the time. It only gets hard for the 20% of edge cases like this one. It doesn't really matter how many successful miles there are if we don't know how well they deal with difficult edge cases.
Slashdot Mirror
That's what I don't get. It would be so simple to certify for these kinds of things before allowing on public roads. Everyone acts like it would be such an impediment. I have to demonstrate I can see without glasses for my license, otherwise I have to drive with glasses. Why does a self-driving car not have to make a basic demonstration of visual/sensor skills before being put on the road? Since self-driving largely relies on the superiority of its sensors for its safety, there should be much higher expectations for sensor testing than a human could actually perform.
I don't even care about your kind of comment any more, because it's totally fine with me if we turn 'doing dumb things' into a capital offense. But what we can't do is allow companies like Uber, of all companies, to make that decision for us. Any way, we've all done dumb things at one point or another. Most of us would not be here if it came to pass. I certainly wouldn't.
So the guy is supposed to say he has cars on the road that would murder someone?
We can know how well humans deal with edge cases by tracking self driving cars and after modelling driving that resembles human driving (meaning mostly diveway to work to grocery store to driveway type stuff). Come up with a criteria for 'difficult event' and log how many occur during a day of normal driving. Compare that statistic to real human accident statistics, extract the gap and there you have the number of accidents prevented by humans.
Darkness can be difficult for humans though.
It's very simple. You think up all the edge cases you can and test. When you're out of ideas, then you put a much safer car on public roads.
Human drivers are very safe compared to Uber, that we know. About other self-driving technologies we know nothing, but I really don't think they are that far ahead of Uber.
These cars are far from making him mainstream. How far behind target is he on Model 3?
Musk acts more like these are hobbies for him. He starts making a model car, then he sees a model rocket in the store that he likes and takes it home and starts that despite not finishing the car. See what I did there?
The first thing regulation should enforce is a set of sensors that are capable of handing these types of events then. Uber obviously isn't, who knows if Waymo is?
I meant 'full' windows managers. I did try DWM. It gets it's speed from pretty much forcing the users to configure it, too much tradeoff.
I have. It is obvious that XFCE gets it's lightness from being minimal.
I call it an edge case because a person dressed in dark standing in a darkly lit section of road with an outcropping behind and choosing to walk into the path of the car is not a common event. Yet it is an event that must be considered, as well as the other million events that contain some novel circumstance.
I've tried a lot of distributions over the years and I can't say I've ever found an interface that felt totally 'light' for linux. I'm not a big fan of OS/X but I can say that it is one thing they got right. Windows even feels light by comparison. I hope it is changing for the better.
So we're going to keep doing this for a million situations? That's a million fatalities. How many edge cases will there be in the real world? Equipment failing is a valid edge case, which brings up my next point. What distance-between interventions misses is the ability of the sensors to operate in all conditions. The camera feeds I saw from the car were terribly dark. Lidar is easily fooled by textured backgrounds. Uber could have avoided this becoming public simply by driving in the day. This tells me not only that it was a weakness that didn't get caught, it was a weakness they didn't even know about.
I'm not talking about accidents though, I'm talking about potentially accident-causing situations. It's a different set of events, because many possible accidents are avoided by humans.
We need self driving vehicles only IF they ever make a dent in the fatality rate. So far I see no evidence at best and I'm not prepared to give them leniency on killing people now based on a mere promise and I don't think anyone should. All evidence right now is that automation isn't coming anywhere close to human safety, which really is quite safe if you take it in perspective of miles driven by humans every day. Being overweight is probably more dangerous and people volunteer for that.
As I said above, 80% of driving is easy, 20% of driving is hard. Not difficult for Waymo to stick to the situations they know they are good at while still being a ticking time bomb on 20% of edge cases. We need full disclosure from these companies, including proof that they can in fact see and recognize ALL objects at ALL times of day and in ALL weather that may be in danger from the car, or be a danger to the car and that they react appropriately.
That sounds great. Why can't corporations get there without killing people on public roads. It's not like there aren't huge testing facilities all over the place. It's not like odd test cases are hard to think up.
80% of driving is easy, 20% of it is full of a million edge cases like this one. Furthermore, edge cases like this are easily avoidable if you have a vested interest in choosing where and when to drive. Waymo could easily rack up 5600 miles and never hit an edge case. We only got elucidated on this one because of a bad test driver and it gave us a window into how bad the situation really is.
This is about saving lives, and is no place for a tit for tat argument. Why can't both sides just try be as safe as FUCKING possible so if one side makes a mistake it is covered by the other side? Is it really that hard to comprehend? Are people really that insensitive that they can't inconvenience themselves by grazing a brake pedal if something unexpected happens up the road?
I'm surprised you didn't get down voted for recommending that.
Another thing I should have added for 2) was the fact that Uber *was* making the requirement for interventions and someone died. So obviously the measurement is bull.
I think that's obvious, but it doesn't change the fact that self driving should be held to a standard where they can accommodate people doing stupid things, otherwise the whole line about saving lines is just blowing smoke. Defensive driving is a 'thing' for good reason and very little of it changes if a computer is driving. If you see something curious, you slow down. Simple?
1) But they have to make something marketable. Then why didn't Uber have a LIDAR array?
2) That's a poor measure. We need to know exactly how many and what type of obstacles the cars are dealing with in those miles. The biggest problem is, self-driving is easy 80% of the time. It only gets hard for the 20% of edge cases like this one. It doesn't really matter how many successful miles there are if we don't know how well they deal with difficult edge cases.